修复跨 Docker 版本灰度镜像校验

This commit is contained in:
2026-10-29 17:16:30 +08:00
parent df688844d3
commit b93a8fccd9
6 changed files with 131 additions and 8 deletions
+3 -1
View File
@@ -24,7 +24,9 @@ done < "$release/release.env"
[[ "$CHAT_IMAGE" =~ ^baiyue-logistics-chat-gray:[a-f0-9]{12}-b[0-9]+$ ]] || exit 2
[[ "$HOST_PORT" =~ ^[0-9]+$ ]] && ((HOST_PORT >= 1024 && HOST_PORT <= 65535)) || exit 2
[[ -d "$CHAT_PROXY_DIR" && -r "$CHAT_TLS_CERT" && -r "$CHAT_TLS_KEY" ]] || exit 2
[[ "$(docker image inspect --format '{{.Id}}' "$CHAT_IMAGE")" == "$CHAT_IMAGE_ID" ]] || exit 2
loaded_id=$(awk -F= '$1 == "CHAT_IMAGE_ID" {print $2}' "$release/loaded-image-ids.env")
[[ "$loaded_id" =~ ^sha256:[a-f0-9]{64}$ ]] || exit 2
[[ "$(docker image inspect --format '{{.Id}}' "$CHAT_IMAGE")" == "$loaded_id" ]] || { echo 'Loaded application image changed.' >&2; exit 2; }
[[ "$(docker image inspect --format '{{index .Config.Labels "org.opencontainers.image.revision"}}' "$CHAT_IMAGE")" == "$SOURCE_COMMIT" ]] || exit 2
openssl x509 -in "$CHAT_TLS_CERT" -checkhost "$CHAT_DOMAIN" -noout >/dev/null
+3 -1
View File
@@ -18,7 +18,9 @@ done < "$release/release.env"
[[ "$CHAT_TLS_CERT" == "$root/tls/config/live/$CHAT_DOMAIN/fullchain.pem" ]] || exit 2
[[ "$CHAT_TLS_KEY" == "$root/tls/config/live/$CHAT_DOMAIN/privkey.pem" ]] || exit 2
[[ "$CHAT_CERTBOT_IMAGE" =~ ^baiyue-logistics-chat-certbot:[a-f0-9]{12}-b[0-9]+$ ]] || exit 2
[[ "$(docker image inspect --format '{{.Id}}' "$CHAT_CERTBOT_IMAGE")" == "$CHAT_CERTBOT_IMAGE_ID" ]] || exit 2
loaded_id=$(awk -F= '$1 == "CERTBOT_IMAGE_ID" {print $2}' "$release/loaded-image-ids.env")
[[ "$loaded_id" =~ ^sha256:[a-f0-9]{64}$ ]] || exit 2
[[ "$(docker image inspect --format '{{.Id}}' "$CHAT_CERTBOT_IMAGE")" == "$loaded_id" ]] || { echo 'Loaded certificate image changed.' >&2; exit 2; }
[[ "$(docker image inspect --format '{{index .Config.Labels "org.opencontainers.image.revision"}}' "$CHAT_CERTBOT_IMAGE")" == "$SOURCE_COMMIT" ]] || exit 2
site="$CHAT_PROXY_DIR/baiyue-logistics-chat-gray.conf"
+33 -3
View File
@@ -40,7 +40,7 @@ release=$2
[[ "$root" == /*/gray-logistics-chat && "$root" != *..* ]] || exit 2
[[ "$release" == "$root"/releases/* ]] || exit 2
[[ "$(id -u)" == 0 ]] || { echo 'The managed gray deployment account must have the required host permissions.' >&2; exit 2; }
for program in docker nginx openssl curl ss systemctl; do command -v "$program" >/dev/null; done
for program in docker nginx openssl curl ss systemctl sha256sum; do command -v "$program" >/dev/null; done
if [[ -e "$root" && ! -f "$root/.project" ]] && [[ -n "$(find "$root" -mindepth 1 -maxdepth 1 -print -quit)" ]]; then
echo 'Refusing to reuse an unowned deployment directory.' >&2
exit 2
@@ -55,8 +55,38 @@ chown 10001:10001 "$root/data" "$root/logs"
install -d -m 700 "$release"
PREPARE
# Transport only tested images built without runtime credentials.
docker save "$CHAT_IMAGE" "$CERTBOT_IMAGE" | gzip | run_ssh "bash -o pipefail -c 'gzip -dc | docker load'"
# Verify the exact exported archive before loading. Classic and containerd image
# stores can report different IDs for the same archive, so retain target IDs too.
docker save "$CHAT_IMAGE" "$CERTBOT_IMAGE" | gzip > "$temporary/images.tar.gz"
archive_sha=$(sha256sum "$temporary/images.tar.gz" | awk '{print $1}')
scp "${ssh_options[@]}" "$temporary/images.tar.gz" "$target:$release/images.tar.gz"
run_ssh bash -s -- "$root" "$release" "$archive_sha" "$CHAT_IMAGE" "$CERTBOT_IMAGE" "$SOURCE_COMMIT" <<'LOAD'
set -Eeuo pipefail
umask 077
root=$1
release=$2
checksum=$3
app_image=$4
certbot_image=$5
source_commit=$6
[[ "$root" == /*/gray-logistics-chat && "$root" != *..* ]] || exit 2
[[ "$release" == "$root"/releases/* && "$release" != *..* ]] || exit 2
[[ "$(cat "$root/.project")" == baiyue-logistics-chat-gray ]] || exit 2
[[ "$checksum" =~ ^[a-f0-9]{64}$ && "$source_commit" =~ ^[a-f0-9]{40}$ ]] || exit 2
[[ "$app_image" =~ ^baiyue-logistics-chat-gray:[a-f0-9]{12}-b[0-9]+$ ]] || exit 2
[[ "$certbot_image" =~ ^baiyue-logistics-chat-certbot:[a-f0-9]{12}-b[0-9]+$ ]] || exit 2
printf '%s %s\n' "$checksum" "$release/images.tar.gz" | sha256sum --check --status
gzip -dc "$release/images.tar.gz" | docker load
for image in "$app_image" "$certbot_image"; do
[[ "$(docker image inspect --format '{{index .Config.Labels "org.opencontainers.image.revision"}}' "$image")" == "$source_commit" ]] || exit 2
done
app_id=$(docker image inspect --format '{{.Id}}' "$app_image")
certbot_id=$(docker image inspect --format '{{.Id}}' "$certbot_image")
[[ "$app_id" =~ ^sha256:[a-f0-9]{64}$ && "$certbot_id" =~ ^sha256:[a-f0-9]{64}$ ]] || exit 2
printf 'ARCHIVE_SHA256=%s\nCHAT_IMAGE_ID=%s\nCERTBOT_IMAGE_ID=%s\n' "$checksum" "$app_id" "$certbot_id" > "$release/loaded-image-ids.env"
rm -f -- "$release/images.tar.gz"
echo 'Image archive checksum and source labels verified; target image IDs recorded.'
LOAD
scp "${ssh_options[@]}" "$temporary/release/"* "$target:$release/"
run_ssh bash "$release/prepare-tls.sh" "$root" "$release"
run_ssh bash "$release/activate.sh" "$root" "$release"
+2 -2
View File
@@ -37,11 +37,11 @@
1. 完成检查并提交代码,创建 `gray-chat-YYYYMMDD-<commit12>` 一类不可变发布标签。不要重写已有标签或强推覆盖远端分支;分支有其他更新时先明确发布基线。
2. 推送标签,在任务 `SOURCE_REF` 选择该标签。Jenkins 按此引用加载流水线并检出代码,镜像记录完整提交 SHA。
3. 流水线检查 Bash 语法,运行 Node 回归,再构建 `test` 目标执行 Python 回归;通过后构建 `production` 应用镜像及 `certbot` 运维镜像。
4. 构建完成后才绑定运行凭据。发布脚本校验路径、域名、源提交和镜像 ID,生成私有配置并发送镜像及发布文件。
4. 构建完成后才绑定运行凭据。发布脚本校验路径、域名和源提交,生成私有配置并发送镜像及发布文件。服务器验证镜像归档的 SHA-256 后加载,记录目标主机的镜像 ID。
5. 首次签发域名证书,等待应用健康,再安装本服务的 Nginx 配置。已有业务的容器和站点配置不参与本次发布。
6. Jenkins 使用验收账号验证公网 HTTPS `/healthz`;之后在浏览器验证 AI 流式回复、上传、转人工和工作台。
应用镜像必须选择 `production` 构建目标;`certbot` 仅用于证书运维。每个构建生成独立镜像标签,接收主机再次核验镜像 ID,不依赖 `latest`。
应用镜像必须选择 `production` 构建目标;`certbot` 仅用于证书运维。每个构建生成独立镜像标签,不依赖 `latest`。Docker 的 classic 与 containerd 镜像存储可能返回不同 ID;跨主机以归档校验和与源提交标签核验传输,激活前再与 `loaded-image-ids.env` 中的目标 ID 比较,拒绝加载后被替换的镜像。归档加载后删除,构建机 ID 保留在 `release.env` 供追溯。
## 数据与失败回退
+11 -1
View File
@@ -66,9 +66,12 @@ class GrayActivationTests(unittest.TestCase):
unrelated.write_text('# unrelated site\n')
config = GrayConfig('chat.example.com', 29013, str(root), str(root / 'releases/new'),
str(proxy), str(certificate), str(private_key), 'a'*40,
'baiyue-logistics-chat-gray:' + 'a'*12 + '-b7', 'sha256:' + 'b'*64, 'sha256:' + 'c'*64)
'baiyue-logistics-chat-gray:' + 'a'*12 + '-b7', 'sha256:' + 'd'*64, 'sha256:' + 'c'*64)
release = Path(config.release)
write_release(release, config, 'fixture-model-key')
# Docker 26 and 29 may assign different IDs after an archive is loaded.
loaded_id = ('e' if scenario == 'image-mismatch' else 'b') * 64
(release / 'loaded-image-ids.env').write_text('CHAT_IMAGE_ID=sha256:' + loaded_id + '\n')
(release / 'site.htpasswd').write_text('new-auth-hash\n')
commands = directory / 'commands'
commands.mkdir()
@@ -105,6 +108,13 @@ class GrayActivationTests(unittest.TestCase):
def test_failed_application_restores_previous_release(self):
self.assert_rolled_back('app-fail')
def test_image_retagged_after_verified_load_is_rejected_before_activation(self):
result, root, proxy, previous, release, old_site, calls = self.run_activation('image-mismatch')
self.assertNotEqual(result.returncode, 0)
self.assertEqual((root / 'current-release').read_text().strip(), str(previous))
self.assertEqual((proxy / 'baiyue-logistics-chat-gray.conf').read_text(), old_site)
self.assertFalse(any(name == 'docker' and 'compose' in args for name, args in calls))
def test_invalid_nginx_configuration_restores_previous_site_and_application(self):
self.assert_rolled_back('proxy-fail')
+79
View File
@@ -0,0 +1,79 @@
"""Verify the real remote image-load guard without contacting Docker or a host."""
import gzip
import hashlib
import os
from pathlib import Path
import shutil
import subprocess
import sys
import tempfile
import unittest
@unittest.skipUnless(os.name == 'posix' and shutil.which('bash'), 'Image transfer requires Linux Bash')
class GrayImageTransferTests(unittest.TestCase):
def run_load(self, scenario):
temporary = tempfile.TemporaryDirectory()
self.addCleanup(temporary.cleanup)
directory = Path(temporary.name)
root = directory / 'gray-logistics-chat'
release = root / 'releases/build'
release.mkdir(parents=True)
(root / '.project').write_text('baiyue-logistics-chat-gray\n')
archive = release / 'images.tar.gz'
archive.write_bytes(gzip.compress(b'fixture-image-archive'))
checksum = hashlib.sha256(archive.read_bytes()).hexdigest()
if scenario == 'corrupt':
archive.write_bytes(archive.read_bytes() + b'changed-in-transit')
commands = directory / 'commands'
commands.mkdir()
docker = commands / 'docker'
docker.write_text('#!' + sys.executable + '\n' + '''import os, pathlib, sys
args = sys.argv[1:]
with pathlib.Path(os.environ['FAKE_LOG']).open('a') as output:
output.write(' '.join(args) + '\\n')
if args == ['load']:
assert sys.stdin.buffer.read() == b'fixture-image-archive'
elif args[:2] == ['image', 'inspect']:
if args[3] == '{{.Id}}':
print('sha256:' + ('b' if '-gray:' in args[-1] else 'c') * 64)
else:
print(('e' if os.environ['FAKE_SCENARIO'] == 'wrong-source' else 'a') * 40)
else:
sys.exit(3)
''')
docker.chmod(0o755)
publish = (Path(__file__).resolve().parents[1] / 'deploy/publish-gray.sh').read_text()
remote = publish.split("<<'LOAD'\n", 1)[1].split('\nLOAD\n', 1)[0]
log = directory / 'commands.log'
environment = dict(os.environ, PATH=str(commands) + os.pathsep + os.environ['PATH'],
FAKE_LOG=str(log), FAKE_SCENARIO=scenario)
result = subprocess.run(['bash', '-s', '--', str(root), str(release), checksum,
'baiyue-logistics-chat-gray:' + 'a' * 12 + '-b7',
'baiyue-logistics-chat-certbot:' + 'a' * 12 + '-b7', 'a' * 40],
input=remote, env=environment, capture_output=True, text=True, timeout=15)
return result, release, log
def test_verified_archive_records_target_ids_and_removes_archive(self):
result, release, log = self.run_load('success')
self.assertEqual(result.returncode, 0, result.stderr)
receipt = (release / 'loaded-image-ids.env').read_text()
self.assertIn('CHAT_IMAGE_ID=sha256:' + 'b' * 64, receipt)
self.assertIn('CERTBOT_IMAGE_ID=sha256:' + 'c' * 64, receipt)
self.assertFalse((release / 'images.tar.gz').exists())
def test_modified_archive_is_rejected_before_docker_load(self):
result, release, log = self.run_load('corrupt')
self.assertNotEqual(result.returncode, 0)
self.assertFalse(log.exists())
self.assertFalse((release / 'loaded-image-ids.env').exists())
def test_wrong_source_cannot_create_a_verified_receipt(self):
result, release, log = self.run_load('wrong-source')
self.assertNotEqual(result.returncode, 0)
self.assertFalse((release / 'loaded-image-ids.env').exists())
if __name__ == '__main__':
unittest.main()